On Tue, Jan 17, 2012 at 02:19:47PM +0100, Marc-André Lureau wrote:
On Tue, Jan 17, 2012 at 2:05 PM, Christophe Fergeau <cfergeau@redhat.com> wrote:
And I forgot to mention that the only remaining use of g_error are in gvir_init_object and gvir_init_config where it's probably acceptable to die if things didn't work as expected.
Not really, it would be better to return an error.
Looking at what gtk+ does, it has gtk_init which is documented to abort, and gtk_init_check which apps can use when they need this. g_type_init has no _check alternative, but it does a few sanity checks at runtime using g_assert. I like the gtk+ approach of aborting in the function most commonly use as it avoids tracking down bugs caused by the _init failing but not being checked by the application developer. I think we should do the same and add _check functions if/when needed (this is me being lazy) Christophe